Forecasting Short Time-Series Tourism Demand with Artificial Intelligence Models

This study examines the forecast accuracy of fuzzy time series and grey theory in predicting annual U.S. tourist arrivals. The performance of the two artificial intelligence (AI) models is compared to that of two simple methods—double moving average and double exponential smoothing. The rigorous testing approach includes a large sample stratified to adequately represent four generic trend patterns: a rolling short-term forecast, a large holdout sample, models fitting with both equal number of years and optimal number of years, and tests of statistical significance using Wilcoxon’s signed-ranks nonparametric test. This study’s findings indicate, in contrast to recent findings, that the complicated models are not likely to generate a more accurate forecast than the simple traditional models. Given the notable cost associated with these AI forecasting methods, our finding of no significant accuracy advantage suggests that tourism forecasters should not rush to adopt these two methods without careful consideration.